net/netfilter/nf_conntrack_proto_tcp.c

   1 /* (C) 1999-2001 Paul `Rusty' Russell
   2  * (C) 2002-2004 Netfilter Core Team <coreteam@netfilter.org>
   3  * (C) 2002-2013 Jozsef Kadlecsik <kadlec@blackhole.kfki.hu>
   4  * (C) 2006-2012 Patrick McHardy <kaber@trash.net>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 as
   8  * published by the Free Software Foundation.
   9  */
  10
  11 #include <linux/types.h>
  12 #include <linux/timer.h>
  13 #include <linux/module.h>
  14 #include <linux/in.h>
  15 #include <linux/tcp.h>
  16 #include <linux/spinlock.h>
  17 #include <linux/skbuff.h>
  18 #include <linux/ipv6.h>
  19 #include <net/ip6_checksum.h>
  20 #include <asm/unaligned.h>
  21
  22 #include <net/tcp.h>
  23
  24 #include <linux/netfilter.h>
  25 #include <linux/netfilter_ipv4.h>
  26 #include <linux/netfilter_ipv6.h>
  27 #include <net/netfilter/nf_conntrack.h>
  28 #include <net/netfilter/nf_conntrack_l4proto.h>
  29 #include <net/netfilter/nf_conntrack_ecache.h>
  30 #include <net/netfilter/nf_conntrack_seqadj.h>
  31 #include <net/netfilter/nf_conntrack_synproxy.h>
  32 #include <net/netfilter/nf_log.h>
  33 #include <net/netfilter/ipv4/nf_conntrack_ipv4.h>
  34 #include <net/netfilter/ipv6/nf_conntrack_ipv6.h>
  35
  36 /* "Be conservative in what you do,
  37     be liberal in what you accept from others."
  38     If it's non-zero, we mark only out of window RST segments as INVALID. */
  39 static int nf_ct_tcp_be_liberal __read_mostly = 0;
  40
  41 /* If it is set to zero, we disable picking up already established
  42    connections. */
  43 static int nf_ct_tcp_loose __read_mostly = 1;
  44
  45 /* Max number of the retransmitted packets without receiving an (acceptable)
  46    ACK from the destination. If this number is reached, a shorter timer
  47    will be started. */
  48 static int nf_ct_tcp_max_retrans __read_mostly = 3;
  49
  50   /* FIXME: Examine ipfilter's timeouts and conntrack transitions more
  51      closely.  They're more complex. --RR */
  52
  53 static const char *const tcp_conntrack_names[] = {
  54         "NONE",
  55         "SYN_SENT",
  56         "SYN_RECV",
  57         "ESTABLISHED",
  58         "FIN_WAIT",
  59         "CLOSE_WAIT",
  60         "LAST_ACK",
  61         "TIME_WAIT",
  62         "CLOSE",
  63         "SYN_SENT2",
  64 };
  65
  66 #define SECS * HZ
  67 #define MINS * 60 SECS
  68 #define HOURS * 60 MINS
  69 #define DAYS * 24 HOURS
  70
  71 static unsigned int tcp_timeouts[TCP_CONNTRACK_TIMEOUT_MAX] __read_mostly = {
  72         [TCP_CONNTRACK_SYN_SENT]        = 2 MINS,
  73         [TCP_CONNTRACK_SYN_RECV]        = 60 SECS,
  74         [TCP_CONNTRACK_ESTABLISHED]     = 5 DAYS,
  75         [TCP_CONNTRACK_FIN_WAIT]        = 2 MINS,
  76         [TCP_CONNTRACK_CLOSE_WAIT]      = 60 SECS,
  77         [TCP_CONNTRACK_LAST_ACK]        = 30 SECS,
  78         [TCP_CONNTRACK_TIME_WAIT]       = 2 MINS,
  79         [TCP_CONNTRACK_CLOSE]           = 10 SECS,
  80         [TCP_CONNTRACK_SYN_SENT2]       = 2 MINS,
  81 /* RFC1122 says the R2 limit should be at least 100 seconds.
  82    Linux uses 15 packets as limit, which corresponds
  83    to ~13-30min depending on RTO. */
  84         [TCP_CONNTRACK_RETRANS]         = 5 MINS,
  85         [TCP_CONNTRACK_UNACK]           = 5 MINS,
  86 };
  87
  88 #define sNO TCP_CONNTRACK_NONE
  89 #define sSS TCP_CONNTRACK_SYN_SENT
  90 #define sSR TCP_CONNTRACK_SYN_RECV
  91 #define sES TCP_CONNTRACK_ESTABLISHED
  92 #define sFW TCP_CONNTRACK_FIN_WAIT
  93 #define sCW TCP_CONNTRACK_CLOSE_WAIT
  94 #define sLA TCP_CONNTRACK_LAST_ACK
  95 #define sTW TCP_CONNTRACK_TIME_WAIT
  96 #define sCL TCP_CONNTRACK_CLOSE
  97 #define sS2 TCP_CONNTRACK_SYN_SENT2
  98 #define sIV TCP_CONNTRACK_MAX
  99 #define sIG TCP_CONNTRACK_IGNORE
 100
 101 /* What TCP flags are set from RST/SYN/FIN/ACK. */
 102 enum tcp_bit_set {
 103         TCP_SYN_SET,
 104         TCP_SYNACK_SET,
 105         TCP_FIN_SET,
 106         TCP_ACK_SET,
 107         TCP_RST_SET,
 108         TCP_NONE_SET,
 109 };
 110
 111 /*
 112  * The TCP state transition table needs a few words...
 113  *
 114  * We are the man in the middle. All the packets go through us
 115  * but might get lost in transit to the destination.
 116  * It is assumed that the destinations can't receive segments
 117  * we haven't seen.
 118  *
 119  * The checked segment is in window, but our windows are *not*
 120  * equivalent with the ones of the sender/receiver. We always
 121  * try to guess the state of the current sender.
 122  *
 123  * The meaning of the states are:
 124  *
 125  * NONE:        initial state
 126  * SYN_SENT:    SYN-only packet seen
 127  * SYN_SENT2:   SYN-only packet seen from reply dir, simultaneous open
 128  * SYN_RECV:    SYN-ACK packet seen
 129  * ESTABLISHED: ACK packet seen
 130  * FIN_WAIT:    FIN packet seen
 131  * CLOSE_WAIT:  ACK seen (after FIN)
 132  * LAST_ACK:    FIN seen (after FIN)
 133  * TIME_WAIT:   last ACK seen
 134  * CLOSE:       closed connection (RST)
 135  *
 136  * Packets marked as IGNORED (sIG):
 137  *      if they may be either invalid or valid
 138  *      and the receiver may send back a connection
 139  *      closing RST or a SYN/ACK.
 140  *
 141  * Packets marked as INVALID (sIV):
 142  *      if we regard them as truly invalid packets
 143  */
 144 static const u8 tcp_conntracks[2][6][TCP_CONNTRACK_MAX] = {
 145         {
 146 /* ORIGINAL */
 147 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 148 /*syn*/    { sSS, sSS, sIG, sIG, sIG, sIG, sIG, sSS, sSS, sS2 },
 149 /*
 150  *      sNO -> sSS      Initialize a new connection
 151  *      sSS -> sSS      Retransmitted SYN
 152  *      sS2 -> sS2      Late retransmitted SYN
 153  *      sSR -> sIG
 154  *      sES -> sIG      Error: SYNs in window outside the SYN_SENT state
 155  *                      are errors. Receiver will reply with RST
 156  *                      and close the connection.
 157  *                      Or we are not in sync and hold a dead connection.
 158  *      sFW -> sIG
 159  *      sCW -> sIG
 160  *      sLA -> sIG
 161  *      sTW -> sSS      Reopened connection (RFC 1122).
 162  *      sCL -> sSS
 163  */
 164 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 165 /*synack*/ { sIV, sIV, sSR, sIV, sIV, sIV, sIV, sIV, sIV, sSR },
 166 /*
 167  *      sNO -> sIV      Too late and no reason to do anything
 168  *      sSS -> sIV      Client can't send SYN and then SYN/ACK
 169  *      sS2 -> sSR      SYN/ACK sent to SYN2 in simultaneous open
 170  *      sSR -> sSR      Late retransmitted SYN/ACK in simultaneous open
 171  *      sES -> sIV      Invalid SYN/ACK packets sent by the client
 172  *      sFW -> sIV
 173  *      sCW -> sIV
 174  *      sLA -> sIV
 175  *      sTW -> sIV
 176  *      sCL -> sIV
 177  */
 178 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 179 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 180 /*
 181  *      sNO -> sIV      Too late and no reason to do anything...
 182  *      sSS -> sIV      Client migth not send FIN in this state:
 183  *                      we enforce waiting for a SYN/ACK reply first.
 184  *      sS2 -> sIV
 185  *      sSR -> sFW      Close started.
 186  *      sES -> sFW
 187  *      sFW -> sLA      FIN seen in both directions, waiting for
 188  *                      the last ACK.
 189  *                      Migth be a retransmitted FIN as well...
 190  *      sCW -> sLA
 191  *      sLA -> sLA      Retransmitted FIN. Remain in the same state.
 192  *      sTW -> sTW
 193  *      sCL -> sCL
 194  */
 195 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 196 /*ack*/    { sES, sIV, sES, sES, sCW, sCW, sTW, sTW, sCL, sIV },
 197 /*
 198  *      sNO -> sES      Assumed.
 199  *      sSS -> sIV      ACK is invalid: we haven't seen a SYN/ACK yet.
 200  *      sS2 -> sIV
 201  *      sSR -> sES      Established state is reached.
 202  *      sES -> sES      :-)
 203  *      sFW -> sCW      Normal close request answered by ACK.
 204  *      sCW -> sCW
 205  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 206  *      sTW -> sTW      Retransmitted last ACK. Remain in the same state.
 207  *      sCL -> sCL
 208  */
 209 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 210 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 211 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 212         },
 213         {
 214 /* REPLY */
 215 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 216 /*syn*/    { sIV, sS2, sIV, sIV, sIV, sIV, sIV, sSS, sIV, sS2 },
 217 /*
 218  *      sNO -> sIV      Never reached.
 219  *      sSS -> sS2      Simultaneous open
 220  *      sS2 -> sS2      Retransmitted simultaneous SYN
 221  *      sSR -> sIV      Invalid SYN packets sent by the server
 222  *      sES -> sIV
 223  *      sFW -> sIV
 224  *      sCW -> sIV
 225  *      sLA -> sIV
 226  *      sTW -> sSS      Reopened connection, but server may have switched role
 227  *      sCL -> sIV
 228  */
 229 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 230 /*synack*/ { sIV, sSR, sIG, sIG, sIG, sIG, sIG, sIG, sIG, sSR },
 231 /*
 232  *      sSS -> sSR      Standard open.
 233  *      sS2 -> sSR      Simultaneous open
 234  *      sSR -> sIG      Retransmitted SYN/ACK, ignore it.
 235  *      sES -> sIG      Late retransmitted SYN/ACK?
 236  *      sFW -> sIG      Might be SYN/ACK answering ignored SYN
 237  *      sCW -> sIG
 238  *      sLA -> sIG
 239  *      sTW -> sIG
 240  *      sCL -> sIG
 241  */
 242 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 243 /*fin*/    { sIV, sIV, sFW, sFW, sLA, sLA, sLA, sTW, sCL, sIV },
 244 /*
 245  *      sSS -> sIV      Server might not send FIN in this state.
 246  *      sS2 -> sIV
 247  *      sSR -> sFW      Close started.
 248  *      sES -> sFW
 249  *      sFW -> sLA      FIN seen in both directions.
 250  *      sCW -> sLA
 251  *      sLA -> sLA      Retransmitted FIN.
 252  *      sTW -> sTW
 253  *      sCL -> sCL
 254  */
 255 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 256 /*ack*/    { sIV, sIG, sSR, sES, sCW, sCW, sTW, sTW, sCL, sIG },
 257 /*
 258  *      sSS -> sIG      Might be a half-open connection.
 259  *      sS2 -> sIG
 260  *      sSR -> sSR      Might answer late resent SYN.
 261  *      sES -> sES      :-)
 262  *      sFW -> sCW      Normal close request answered by ACK.
 263  *      sCW -> sCW
 264  *      sLA -> sTW      Last ACK detected (RFC5961 challenged)
 265  *      sTW -> sTW      Retransmitted last ACK.
 266  *      sCL -> sCL
 267  */
 268 /*           sNO, sSS, sSR, sES, sFW, sCW, sLA, sTW, sCL, sS2   */
 269 /*rst*/    { sIV, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL, sCL },
 270 /*none*/   { sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV, sIV }
 271         }
 272 };
 273
 274 static inline struct nf_tcp_net *tcp_pernet(struct net *net)
 275 {
 276         return &net->ct.nf_ct_proto.tcp;
 277 }
 278
 279 static bool tcp_pkt_to_tuple(const struct sk_buff *skb, unsigned int dataoff,
 280                              struct net *net, struct nf_conntrack_tuple *tuple)
 281 {
 282         const struct tcphdr *hp;
 283         struct tcphdr _hdr;
 284
 285         /* Actually only need first 4 bytes to get ports. */
 286         hp = skb_header_pointer(skb, dataoff, 4, &_hdr);
 287         if (hp == NULL)
 288                 return false;
 289
 290         tuple->src.u.tcp.port = hp->source;
 291         tuple->dst.u.tcp.port = hp->dest;
 292
 293         return true;
 294 }
 295
 296 static bool tcp_invert_tuple(struct nf_conntrack_tuple *tuple,
 297                              const struct nf_conntrack_tuple *orig)
 298 {
 299         tuple->src.u.tcp.port = orig->dst.u.tcp.port;
 300         tuple->dst.u.tcp.port = orig->src.u.tcp.port;
 301         return true;
 302 }
 303
 304 #ifdef CONFIG_NF_CONNTRACK_PROCFS
 305 /* Print out the private part of the conntrack. */
 306 static void tcp_print_conntrack(struct seq_file *s, struct nf_conn *ct)
 307 {
 308         seq_printf(s, "%s ", tcp_conntrack_names[ct->proto.tcp.state]);
 309 }
 310 #endif
 311
 312 static unsigned int get_conntrack_index(const struct tcphdr *tcph)
 313 {
 314         if (tcph->rst) return TCP_RST_SET;
 315         else if (tcph->syn) return (tcph->ack ? TCP_SYNACK_SET : TCP_SYN_SET);
 316         else if (tcph->fin) return TCP_FIN_SET;
 317         else if (tcph->ack) return TCP_ACK_SET;
 318         else return TCP_NONE_SET;
 319 }
 320
 321 /* TCP connection tracking based on 'Real Stateful TCP Packet Filtering
 322    in IP Filter' by Guido van Rooij.
 323
 324    http://www.sane.nl/events/sane2000/papers.html
 325    http://www.darkart.com/mirrors/www.obfuscation.org/ipf/
 326
 327    The boundaries and the conditions are changed according to RFC793:
 328    the packet must intersect the window (i.e. segments may be
 329    after the right or before the left edge) and thus receivers may ACK
 330    segments after the right edge of the window.
 331
 332         td_maxend = max(sack + max(win,1)) seen in reply packets
 333         td_maxwin = max(max(win, 1)) + (sack - ack) seen in sent packets
 334         td_maxwin += seq + len - sender.td_maxend
 335                         if seq + len > sender.td_maxend
 336         td_end    = max(seq + len) seen in sent packets
 337
 338    I.   Upper bound for valid data:     seq <= sender.td_maxend
 339    II.  Lower bound for valid data:     seq + len >= sender.td_end - receiver.td_maxwin
 340    III. Upper bound for valid (s)ack:   sack <= receiver.td_end
 341    IV.  Lower bound for valid (s)ack:   sack >= receiver.td_end - MAXACKWINDOW
 342
 343    where sack is the highest right edge of sack block found in the packet
 344    or ack in the case of packet without SACK option.
 345
 346    The upper bound limit for a valid (s)ack is not ignored -
 347    we doesn't have to deal with fragments.
 348 */
 349
 350 static inline __u32 segment_seq_plus_len(__u32 seq,
 351                                          size_t len,
 352                                          unsigned int dataoff,
 353                                          const struct tcphdr *tcph)
 354 {
 355         /* XXX Should I use payload length field in IP/IPv6 header ?
 356          * - YK */
 357         return (seq + len - dataoff - tcph->doff*4
 358                 + (tcph->syn ? 1 : 0) + (tcph->fin ? 1 : 0));
 359 }
 360
 361 /* Fixme: what about big packets? */
 362 #define MAXACKWINCONST                  66000
 363 #define MAXACKWINDOW(sender)                                            \
 364         ((sender)->td_maxwin > MAXACKWINCONST ? (sender)->td_maxwin     \
 365                                               : MAXACKWINCONST)
 366
 367 /*
 368  * Simplified tcp_parse_options routine from tcp_input.c
 369  */
 370 static void tcp_options(const struct sk_buff *skb,
 371                         unsigned int dataoff,
 372                         const struct tcphdr *tcph,
 373                         struct ip_ct_tcp_state *state)
 374 {
 375         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 376         const unsigned char *ptr;
 377         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 378
 379         if (!length)
 380                 return;
 381
 382         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 383                                  length, buff);
 384         BUG_ON(ptr == NULL);
 385
 386         state->td_scale =
 387         state->flags = 0;
 388
 389         while (length > 0) {
 390                 int opcode=*ptr++;
 391                 int opsize;
 392
 393                 switch (opcode) {
 394                 case TCPOPT_EOL:
 395                         return;
 396                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 397                         length--;
 398                         continue;
 399                 default:
 400                         if (length < 2)
 401                                 return;
 402                         opsize=*ptr++;
 403                         if (opsize < 2) /* "silly options" */
 404                                 return;
 405                         if (opsize > length)
 406                                 return; /* don't parse partial options */
 407
 408                         if (opcode == TCPOPT_SACK_PERM
 409                             && opsize == TCPOLEN_SACK_PERM)
 410                                 state->flags |= IP_CT_TCP_FLAG_SACK_PERM;
 411                         else if (opcode == TCPOPT_WINDOW
 412                                  && opsize == TCPOLEN_WINDOW) {
 413                                 state->td_scale = *(u_int8_t *)ptr;
 414
 415                                 if (state->td_scale > TCP_MAX_WSCALE)
 416                                         state->td_scale = TCP_MAX_WSCALE;
 417
 418                                 state->flags |=
 419                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 420                         }
 421                         ptr += opsize - 2;
 422                         length -= opsize;
 423                 }
 424         }
 425 }
 426
 427 static void tcp_sack(const struct sk_buff *skb, unsigned int dataoff,
 428                      const struct tcphdr *tcph, __u32 *sack)
 429 {
 430         unsigned char buff[(15 * 4) - sizeof(struct tcphdr)];
 431         const unsigned char *ptr;
 432         int length = (tcph->doff*4) - sizeof(struct tcphdr);
 433         __u32 tmp;
 434
 435         if (!length)
 436                 return;
 437
 438         ptr = skb_header_pointer(skb, dataoff + sizeof(struct tcphdr),
 439                                  length, buff);
 440         BUG_ON(ptr == NULL);
 441
 442         /* Fast path for timestamp-only option */
 443         if (length == TCPOLEN_TSTAMP_ALIGNED
 444             && *(__be32 *)ptr == htonl((TCPOPT_NOP << 24)
 445                                        | (TCPOPT_NOP << 16)
 446                                        | (TCPOPT_TIMESTAMP << 8)
 447                                        | TCPOLEN_TIMESTAMP))
 448                 return;
 449
 450         while (length > 0) {
 451                 int opcode = *ptr++;
 452                 int opsize, i;
 453
 454                 switch (opcode) {
 455                 case TCPOPT_EOL:
 456                         return;
 457                 case TCPOPT_NOP:        /* Ref: RFC 793 section 3.1 */
 458                         length--;
 459                         continue;
 460                 default:
 461                         if (length < 2)
 462                                 return;
 463                         opsize = *ptr++;
 464                         if (opsize < 2) /* "silly options" */
 465                                 return;
 466                         if (opsize > length)
 467                                 return; /* don't parse partial options */
 468
 469                         if (opcode == TCPOPT_SACK
 470                             && opsize >= (TCPOLEN_SACK_BASE
 471                                           + TCPOLEN_SACK_PERBLOCK)
 472                             && !((opsize - TCPOLEN_SACK_BASE)
 473                                  % TCPOLEN_SACK_PERBLOCK)) {
 474                                 for (i = 0;
 475                                      i < (opsize - TCPOLEN_SACK_BASE);
 476                                      i += TCPOLEN_SACK_PERBLOCK) {
 477                                         tmp = get_unaligned_be32((__be32 *)(ptr+i)+1);
 478
 479                                         if (after(tmp, *sack))
 480                                                 *sack = tmp;
 481                                 }
 482                                 return;
 483                         }
 484                         ptr += opsize - 2;
 485                         length -= opsize;
 486                 }
 487         }
 488 }
 489
 490 static bool tcp_in_window(const struct nf_conn *ct,
 491                           struct ip_ct_tcp *state,
 492                           enum ip_conntrack_dir dir,
 493                           unsigned int index,
 494                           const struct sk_buff *skb,
 495                           unsigned int dataoff,
 496                           const struct tcphdr *tcph)
 497 {
 498         struct net *net = nf_ct_net(ct);
 499         struct nf_tcp_net *tn = tcp_pernet(net);
 500         struct ip_ct_tcp_state *sender = &state->seen[dir];
 501         struct ip_ct_tcp_state *receiver = &state->seen[!dir];
 502         const struct nf_conntrack_tuple *tuple = &ct->tuplehash[dir].tuple;
 503         __u32 seq, ack, sack, end, win, swin;
 504         s32 receiver_offset;
 505         bool res, in_recv_win;
 506
 507         /*
 508          * Get the required data from the packet.
 509          */
 510         seq = ntohl(tcph->seq);
 511         ack = sack = ntohl(tcph->ack_seq);
 512         win = ntohs(tcph->window);
 513         end = segment_seq_plus_len(seq, skb->len, dataoff, tcph);
 514
 515         if (receiver->flags & IP_CT_TCP_FLAG_SACK_PERM)
 516                 tcp_sack(skb, dataoff, tcph, &sack);
 517
 518         /* Take into account NAT sequence number mangling */
 519         receiver_offset = nf_ct_seq_offset(ct, !dir, ack - 1);
 520         ack -= receiver_offset;
 521         sack -= receiver_offset;
 522
 523         pr_debug("tcp_in_window: START\n");
 524         pr_debug("tcp_in_window: ");
 525         nf_ct_dump_tuple(tuple);
 526         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 527                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 528         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 529                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 530                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 531                  sender->td_scale,
 532                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 533                  receiver->td_scale);
 534
 535         if (sender->td_maxwin == 0) {
 536                 /*
 537                  * Initialize sender data.
 538                  */
 539                 if (tcph->syn) {
 540                         /*
 541                          * SYN-ACK in reply to a SYN
 542                          * or SYN from reply direction in simultaneous open.
 543                          */
 544                         sender->td_end =
 545                         sender->td_maxend = end;
 546                         sender->td_maxwin = (win == 0 ? 1 : win);
 547
 548                         tcp_options(skb, dataoff, tcph, sender);
 549                         /*
 550                          * RFC 1323:
 551                          * Both sides must send the Window Scale option
 552                          * to enable window scaling in either direction.
 553                          */
 554                         if (!(sender->flags & IP_CT_TCP_FLAG_WINDOW_SCALE
 555                               && receiver->flags & IP_CT_TCP_FLAG_WINDOW_SCALE))
 556                                 sender->td_scale =
 557                                 receiver->td_scale = 0;
 558                         if (!tcph->ack)
 559                                 /* Simultaneous open */
 560                                 return true;
 561                 } else {
 562                         /*
 563                          * We are in the middle of a connection,
 564                          * its history is lost for us.
 565                          * Let's try to use the data from the packet.
 566                          */
 567                         sender->td_end = end;
 568                         swin = win << sender->td_scale;
 569                         sender->td_maxwin = (swin == 0 ? 1 : swin);
 570                         sender->td_maxend = end + sender->td_maxwin;
 571                         /*
 572                          * We haven't seen traffic in the other direction yet
 573                          * but we have to tweak window tracking to pass III
 574                          * and IV until that happens.
 575                          */
 576                         if (receiver->td_maxwin == 0)
 577                                 receiver->td_end = receiver->td_maxend = sack;
 578                 }
 579         } else if (((state->state == TCP_CONNTRACK_SYN_SENT
 580                      && dir == IP_CT_DIR_ORIGINAL)
 581                    || (state->state == TCP_CONNTRACK_SYN_RECV
 582                      && dir == IP_CT_DIR_REPLY))
 583                    && after(end, sender->td_end)) {
 584                 /*
 585                  * RFC 793: "if a TCP is reinitialized ... then it need
 586                  * not wait at all; it must only be sure to use sequence
 587                  * numbers larger than those recently used."
 588                  */
 589                 sender->td_end =
 590                 sender->td_maxend = end;
 591                 sender->td_maxwin = (win == 0 ? 1 : win);
 592
 593                 tcp_options(skb, dataoff, tcph, sender);
 594         }
 595
 596         if (!(tcph->ack)) {
 597                 /*
 598                  * If there is no ACK, just pretend it was set and OK.
 599                  */
 600                 ack = sack = receiver->td_end;
 601         } else if (((tcp_flag_word(tcph) & (TCP_FLAG_ACK|TCP_FLAG_RST)) ==
 602                     (TCP_FLAG_ACK|TCP_FLAG_RST))
 603                    && (ack == 0)) {
 604                 /*
 605                  * Broken TCP stacks, that set ACK in RST packets as well
 606                  * with zero ack value.
 607                  */
 608                 ack = sack = receiver->td_end;
 609         }
 610
 611         if (tcph->rst && seq == 0 && state->state == TCP_CONNTRACK_SYN_SENT)
 612                 /*
 613                  * RST sent answering SYN.
 614                  */
 615                 seq = end = sender->td_end;
 616
 617         pr_debug("tcp_in_window: ");
 618         nf_ct_dump_tuple(tuple);
 619         pr_debug("seq=%u ack=%u+(%d) sack=%u+(%d) win=%u end=%u\n",
 620                  seq, ack, receiver_offset, sack, receiver_offset, win, end);
 621         pr_debug("tcp_in_window: sender end=%u maxend=%u maxwin=%u scale=%i "
 622                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
 623                  sender->td_end, sender->td_maxend, sender->td_maxwin,
 624                  sender->td_scale,
 625                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
 626                  receiver->td_scale);
 627
 628         /* Is the ending sequence in the receive window (if available)? */
 629         in_recv_win = !receiver->td_maxwin ||
 630                       after(end, sender->td_end - receiver->td_maxwin - 1);
 631
 632         pr_debug("tcp_in_window: I=%i II=%i III=%i IV=%i\n",
 633                  before(seq, sender->td_maxend + 1),
 634                  (in_recv_win ? 1 : 0),
 635                  before(sack, receiver->td_end + 1),
 636                  after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1));
 637
 638         if (before(seq, sender->td_maxend + 1) &&
 639             in_recv_win &&
 640             before(sack, receiver->td_end + 1) &&
 641             after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1)) {
 642                 /*
 643                  * Take into account window scaling (RFC 1323).
 644                  */
 645                 if (!tcph->syn)
 646                         win <<= sender->td_scale;
 647
 648                 /*
 649                  * Update sender data.
 650                  */
 651                 swin = win + (sack - ack);
 652                 if (sender->td_maxwin < swin)
 653                         sender->td_maxwin = swin;
 654                 if (after(end, sender->td_end)) {
 655                         sender->td_end = end;
 656                         sender->flags |= IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 657                 }
 658                 if (tcph->ack) {
 659                         if (!(sender->flags & IP_CT_TCP_FLAG_MAXACK_SET)) {
 660                                 sender->td_maxack = ack;
 661                                 sender->flags |= IP_CT_TCP_FLAG_MAXACK_SET;
 662                         } else if (after(ack, sender->td_maxack))
 663                                 sender->td_maxack = ack;
 664                 }
 665
 666                 /*
 667                  * Update receiver data.
 668                  */
 669                 if (receiver->td_maxwin != 0 && after(end, sender->td_maxend))
 670                         receiver->td_maxwin += end - sender->td_maxend;
 671                 if (after(sack + win, receiver->td_maxend - 1)) {
 672                         receiver->td_maxend = sack + win;
 673                         if (win == 0)
 674                                 receiver->td_maxend++;
 675                 }
 676                 if (ack == receiver->td_end)
 677                         receiver->flags &= ~IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED;
 678
 679                 /*
 680                  * Check retransmissions.
 681                  */
 682                 if (index == TCP_ACK_SET) {
 683                         if (state->last_dir == dir
 684                             && state->last_seq == seq
 685                             && state->last_ack == ack
 686                             && state->last_end == end
 687                             && state->last_win == win)
 688                                 state->retrans++;
 689                         else {
 690                                 state->last_dir = dir;
 691                                 state->last_seq = seq;
 692                                 state->last_ack = ack;
 693                                 state->last_end = end;
 694                                 state->last_win = win;
 695                                 state->retrans = 0;
 696                         }
 697                 }
 698                 res = true;
 699         } else {
 700                 res = false;
 701                 if (sender->flags & IP_CT_TCP_FLAG_BE_LIBERAL ||
 702                     tn->tcp_be_liberal)
 703                         res = true;
 704                 if (!res) {
 705                         nf_ct_l4proto_log_invalid(skb, ct,
 706                         "%s",
 707                         before(seq, sender->td_maxend + 1) ?
 708                         in_recv_win ?
 709                         before(sack, receiver->td_end + 1) ?
 710                         after(sack, receiver->td_end - MAXACKWINDOW(sender) - 1) ? "BUG"
 711                         : "ACK is under the lower bound (possible overly delayed ACK)"
 712                         : "ACK is over the upper bound (ACKed data not seen yet)"
 713                         : "SEQ is under the lower bound (already ACKed data retransmitted)"
 714                         : "SEQ is over the upper bound (over the window of the receiver)");
 715                 }
 716         }
 717
 718         pr_debug("tcp_in_window: res=%u sender end=%u maxend=%u maxwin=%u "
 719                  "receiver end=%u maxend=%u maxwin=%u\n",
 720                  res, sender->td_end, sender->td_maxend, sender->td_maxwin,
 721                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin);
 722
 723         return res;
 724 }
 725
 726 /* table of valid flag combinations - PUSH, ECE and CWR are always valid */
 727 static const u8 tcp_valid_flags[(TCPHDR_FIN|TCPHDR_SYN|TCPHDR_RST|TCPHDR_ACK|
 728                                  TCPHDR_URG) + 1] =
 729 {
 730         [TCPHDR_SYN]                            = 1,
 731         [TCPHDR_SYN|TCPHDR_URG]                 = 1,
 732         [TCPHDR_SYN|TCPHDR_ACK]                 = 1,
 733         [TCPHDR_RST]                            = 1,
 734         [TCPHDR_RST|TCPHDR_ACK]                 = 1,
 735         [TCPHDR_FIN|TCPHDR_ACK]                 = 1,
 736         [TCPHDR_FIN|TCPHDR_ACK|TCPHDR_URG]      = 1,
 737         [TCPHDR_ACK]                            = 1,
 738         [TCPHDR_ACK|TCPHDR_URG]                 = 1,
 739 };
 740
 741 static void tcp_error_log(const struct sk_buff *skb, struct net *net,
 742                           u8 pf, const char *msg)
 743 {
 744         nf_l4proto_log_invalid(skb, net, pf, IPPROTO_TCP, "%s", msg);
 745 }
 746
 747 /* Protect conntrack agaist broken packets. Code taken from ipt_unclean.c.  */
 748 static int tcp_error(struct net *net, struct nf_conn *tmpl,
 749                      struct sk_buff *skb,
 750                      unsigned int dataoff,
 751                      u_int8_t pf,
 752                      unsigned int hooknum)
 753 {
 754         const struct tcphdr *th;
 755         struct tcphdr _tcph;
 756         unsigned int tcplen = skb->len - dataoff;
 757         u_int8_t tcpflags;
 758
 759         /* Smaller that minimal TCP header? */
 760         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 761         if (th == NULL) {
 762                 tcp_error_log(skb, net, pf, "short packet");
 763                 return -NF_ACCEPT;
 764         }
 765
 766         /* Not whole TCP header or malformed packet */
 767         if (th->doff*4 < sizeof(struct tcphdr) || tcplen < th->doff*4) {
 768                 tcp_error_log(skb, net, pf, "truncated packet");
 769                 return -NF_ACCEPT;
 770         }
 771
 772         /* Checksum invalid? Ignore.
 773          * We skip checking packets on the outgoing path
 774          * because the checksum is assumed to be correct.
 775          */
 776         /* FIXME: Source route IP option packets --RR */
 777         if (net->ct.sysctl_checksum && hooknum == NF_INET_PRE_ROUTING &&
 778             nf_checksum(skb, hooknum, dataoff, IPPROTO_TCP, pf)) {
 779                 tcp_error_log(skb, net, pf, "bad checksum");
 780                 return -NF_ACCEPT;
 781         }
 782
 783         /* Check TCP flags. */
 784         tcpflags = (tcp_flag_byte(th) & ~(TCPHDR_ECE|TCPHDR_CWR|TCPHDR_PSH));
 785         if (!tcp_valid_flags[tcpflags]) {
 786                 tcp_error_log(skb, net, pf, "invalid tcp flag combination");
 787                 return -NF_ACCEPT;
 788         }
 789
 790         return NF_ACCEPT;
 791 }
 792
 793 static unsigned int *tcp_get_timeouts(struct net *net)
 794 {
 795         return tcp_pernet(net)->timeouts;
 796 }
 797
 798 /* Returns verdict for packet, or -1 for invalid. */
 799 static int tcp_packet(struct nf_conn *ct,
 800                       const struct sk_buff *skb,
 801                       unsigned int dataoff,
 802                       enum ip_conntrack_info ctinfo,
 803                       unsigned int *timeouts)
 804 {
 805         struct net *net = nf_ct_net(ct);
 806         struct nf_tcp_net *tn = tcp_pernet(net);
 807         struct nf_conntrack_tuple *tuple;
 808         enum tcp_conntrack new_state, old_state;
 809         enum ip_conntrack_dir dir;
 810         const struct tcphdr *th;
 811         struct tcphdr _tcph;
 812         unsigned long timeout;
 813         unsigned int index;
 814
 815         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
 816         BUG_ON(th == NULL);
 817
 818         spin_lock_bh(&ct->lock);
 819         old_state = ct->proto.tcp.state;
 820         dir = CTINFO2DIR(ctinfo);
 821         index = get_conntrack_index(th);
 822         new_state = tcp_conntracks[dir][index][old_state];
 823         tuple = &ct->tuplehash[dir].tuple;
 824
 825         switch (new_state) {
 826         case TCP_CONNTRACK_SYN_SENT:
 827                 if (old_state < TCP_CONNTRACK_TIME_WAIT)
 828                         break;
 829                 /* RFC 1122: "When a connection is closed actively,
 830                  * it MUST linger in TIME-WAIT state for a time 2xMSL
 831                  * (Maximum Segment Lifetime). However, it MAY accept
 832                  * a new SYN from the remote TCP to reopen the connection
 833                  * directly from TIME-WAIT state, if..."
 834                  * We ignore the conditions because we are in the
 835                  * TIME-WAIT state anyway.
 836                  *
 837                  * Handle aborted connections: we and the server
 838                  * think there is an existing connection but the client
 839                  * aborts it and starts a new one.
 840                  */
 841                 if (((ct->proto.tcp.seen[dir].flags
 842                       | ct->proto.tcp.seen[!dir].flags)
 843                      & IP_CT_TCP_FLAG_CLOSE_INIT)
 844                     || (ct->proto.tcp.last_dir == dir
 845                         && ct->proto.tcp.last_index == TCP_RST_SET)) {
 846                         /* Attempt to reopen a closed/aborted connection.
 847                          * Delete this connection and look up again. */
 848                         spin_unlock_bh(&ct->lock);
 849
 850                         /* Only repeat if we can actually remove the timer.
 851                          * Destruction may already be in progress in process
 852                          * context and we must give it a chance to terminate.
 853                          */
 854                         if (nf_ct_kill(ct))
 855                                 return -NF_REPEAT;
 856                         return NF_DROP;
 857                 }
 858                 /* Fall through */
 859         case TCP_CONNTRACK_IGNORE:
 860                 /* Ignored packets:
 861                  *
 862                  * Our connection entry may be out of sync, so ignore
 863                  * packets which may signal the real connection between
 864                  * the client and the server.
 865                  *
 866                  * a) SYN in ORIGINAL
 867                  * b) SYN/ACK in REPLY
 868                  * c) ACK in reply direction after initial SYN in original.
 869                  *
 870                  * If the ignored packet is invalid, the receiver will send
 871                  * a RST we'll catch below.
 872                  */
 873                 if (index == TCP_SYNACK_SET
 874                     && ct->proto.tcp.last_index == TCP_SYN_SET
 875                     && ct->proto.tcp.last_dir != dir
 876                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 877                         /* b) This SYN/ACK acknowledges a SYN that we earlier
 878                          * ignored as invalid. This means that the client and
 879                          * the server are both in sync, while the firewall is
 880                          * not. We get in sync from the previously annotated
 881                          * values.
 882                          */
 883                         old_state = TCP_CONNTRACK_SYN_SENT;
 884                         new_state = TCP_CONNTRACK_SYN_RECV;
 885                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_end =
 886                                 ct->proto.tcp.last_end;
 887                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxend =
 888                                 ct->proto.tcp.last_end;
 889                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_maxwin =
 890                                 ct->proto.tcp.last_win == 0 ?
 891                                         1 : ct->proto.tcp.last_win;
 892                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].td_scale =
 893                                 ct->proto.tcp.last_wscale;
 894                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 895                         ct->proto.tcp.seen[ct->proto.tcp.last_dir].flags =
 896                                 ct->proto.tcp.last_flags;
 897                         memset(&ct->proto.tcp.seen[dir], 0,
 898                                sizeof(struct ip_ct_tcp_state));
 899                         break;
 900                 }
 901                 ct->proto.tcp.last_index = index;
 902                 ct->proto.tcp.last_dir = dir;
 903                 ct->proto.tcp.last_seq = ntohl(th->seq);
 904                 ct->proto.tcp.last_end =
 905                     segment_seq_plus_len(ntohl(th->seq), skb->len, dataoff, th);
 906                 ct->proto.tcp.last_win = ntohs(th->window);
 907
 908                 /* a) This is a SYN in ORIGINAL. The client and the server
 909                  * may be in sync but we are not. In that case, we annotate
 910                  * the TCP options and let the packet go through. If it is a
 911                  * valid SYN packet, the server will reply with a SYN/ACK, and
 912                  * then we'll get in sync. Otherwise, the server potentially
 913                  * responds with a challenge ACK if implementing RFC5961.
 914                  */
 915                 if (index == TCP_SYN_SET && dir == IP_CT_DIR_ORIGINAL) {
 916                         struct ip_ct_tcp_state seen = {};
 917
 918                         ct->proto.tcp.last_flags =
 919                         ct->proto.tcp.last_wscale = 0;
 920                         tcp_options(skb, dataoff, th, &seen);
 921                         if (seen.flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
 922                                 ct->proto.tcp.last_flags |=
 923                                         IP_CT_TCP_FLAG_WINDOW_SCALE;
 924                                 ct->proto.tcp.last_wscale = seen.td_scale;
 925                         }
 926                         if (seen.flags & IP_CT_TCP_FLAG_SACK_PERM) {
 927                                 ct->proto.tcp.last_flags |=
 928                                         IP_CT_TCP_FLAG_SACK_PERM;
 929                         }
 930                         /* Mark the potential for RFC5961 challenge ACK,
 931                          * this pose a special problem for LAST_ACK state
 932                          * as ACK is intrepretated as ACKing last FIN.
 933                          */
 934                         if (old_state == TCP_CONNTRACK_LAST_ACK)
 935                                 ct->proto.tcp.last_flags |=
 936                                         IP_CT_EXP_CHALLENGE_ACK;
 937                 }
 938                 spin_unlock_bh(&ct->lock);
 939                 nf_ct_l4proto_log_invalid(skb, ct, "invalid packet ignored in "
 940                                           "state %s ", tcp_conntrack_names[old_state]);
 941                 return NF_ACCEPT;
 942         case TCP_CONNTRACK_MAX:
 943                 /* Special case for SYN proxy: when the SYN to the server or
 944                  * the SYN/ACK from the server is lost, the client may transmit
 945                  * a keep-alive packet while in SYN_SENT state. This needs to
 946                  * be associated with the original conntrack entry in order to
 947                  * generate a new SYN with the correct sequence number.
 948                  */
 949                 if (nfct_synproxy(ct) && old_state == TCP_CONNTRACK_SYN_SENT &&
 950                     index == TCP_ACK_SET && dir == IP_CT_DIR_ORIGINAL &&
 951                     ct->proto.tcp.last_dir == IP_CT_DIR_ORIGINAL &&
 952                     ct->proto.tcp.seen[dir].td_end - 1 == ntohl(th->seq)) {
 953                         pr_debug("nf_ct_tcp: SYN proxy client keep alive\n");
 954                         spin_unlock_bh(&ct->lock);
 955                         return NF_ACCEPT;
 956                 }
 957
 958                 /* Invalid packet */
 959                 pr_debug("nf_ct_tcp: Invalid dir=%i index=%u ostate=%u\n",
 960                          dir, get_conntrack_index(th), old_state);
 961                 spin_unlock_bh(&ct->lock);
 962                 nf_ct_l4proto_log_invalid(skb, ct, "invalid state");
 963                 return -NF_ACCEPT;
 964         case TCP_CONNTRACK_TIME_WAIT:
 965                 /* RFC5961 compliance cause stack to send "challenge-ACK"
 966                  * e.g. in response to spurious SYNs.  Conntrack MUST
 967                  * not believe this ACK is acking last FIN.
 968                  */
 969                 if (old_state == TCP_CONNTRACK_LAST_ACK &&
 970                     index == TCP_ACK_SET &&
 971                     ct->proto.tcp.last_dir != dir &&
 972                     ct->proto.tcp.last_index == TCP_SYN_SET &&
 973                     (ct->proto.tcp.last_flags & IP_CT_EXP_CHALLENGE_ACK)) {
 974                         /* Detected RFC5961 challenge ACK */
 975                         ct->proto.tcp.last_flags &= ~IP_CT_EXP_CHALLENGE_ACK;
 976                         spin_unlock_bh(&ct->lock);
 977                         nf_ct_l4proto_log_invalid(skb, ct, "challenge-ack ignored");
 978                         return NF_ACCEPT; /* Don't change state */
 979                 }
 980                 break;
 981         case TCP_CONNTRACK_CLOSE:
 982                 if (index == TCP_RST_SET
 983                     && (ct->proto.tcp.seen[!dir].flags & IP_CT_TCP_FLAG_MAXACK_SET)
 984                     && before(ntohl(th->seq), ct->proto.tcp.seen[!dir].td_maxack)) {
 985                         /* Invalid RST  */
 986                         spin_unlock_bh(&ct->lock);
 987                         nf_ct_l4proto_log_invalid(skb, ct, "invalid rst");
 988                         return -NF_ACCEPT;
 989                 }
 990                 if (index == TCP_RST_SET
 991                     && ((test_bit(IPS_SEEN_REPLY_BIT, &ct->status)
 992                          && ct->proto.tcp.last_index == TCP_SYN_SET)
 993                         || (!test_bit(IPS_ASSURED_BIT, &ct->status)
 994                             && ct->proto.tcp.last_index == TCP_ACK_SET))
 995                     && ntohl(th->ack_seq) == ct->proto.tcp.last_end) {
 996                         /* RST sent to invalid SYN or ACK we had let through
 997                          * at a) and c) above:
 998                          *
 999                          * a) SYN was in window then
1000                          * c) we hold a half-open connection.
1001                          *
1002                          * Delete our connection entry.
1003                          * We skip window checking, because packet might ACK
1004                          * segments we ignored. */
1005                         goto in_window;
1006                 }
1007                 /* Just fall through */
1008         default:
1009                 /* Keep compilers happy. */
1010                 break;
1011         }
1012
1013         if (!tcp_in_window(ct, &ct->proto.tcp, dir, index,
1014                            skb, dataoff, th)) {
1015                 spin_unlock_bh(&ct->lock);
1016                 return -NF_ACCEPT;
1017         }
1018      in_window:
1019         /* From now on we have got in-window packets */
1020         ct->proto.tcp.last_index = index;
1021         ct->proto.tcp.last_dir = dir;
1022
1023         pr_debug("tcp_conntracks: ");
1024         nf_ct_dump_tuple(tuple);
1025         pr_debug("syn=%i ack=%i fin=%i rst=%i old=%i new=%i\n",
1026                  (th->syn ? 1 : 0), (th->ack ? 1 : 0),
1027                  (th->fin ? 1 : 0), (th->rst ? 1 : 0),
1028                  old_state, new_state);
1029
1030         ct->proto.tcp.state = new_state;
1031         if (old_state != new_state
1032             && new_state == TCP_CONNTRACK_FIN_WAIT)
1033                 ct->proto.tcp.seen[dir].flags |= IP_CT_TCP_FLAG_CLOSE_INIT;
1034
1035         if (ct->proto.tcp.retrans >= tn->tcp_max_retrans &&
1036             timeouts[new_state] > timeouts[TCP_CONNTRACK_RETRANS])
1037                 timeout = timeouts[TCP_CONNTRACK_RETRANS];
1038         else if ((ct->proto.tcp.seen[0].flags | ct->proto.tcp.seen[1].flags) &
1039                  IP_CT_TCP_FLAG_DATA_UNACKNOWLEDGED &&
1040                  timeouts[new_state] > timeouts[TCP_CONNTRACK_UNACK])
1041                 timeout = timeouts[TCP_CONNTRACK_UNACK];
1042         else
1043                 timeout = timeouts[new_state];
1044         spin_unlock_bh(&ct->lock);
1045
1046         if (new_state != old_state)
1047                 nf_conntrack_event_cache(IPCT_PROTOINFO, ct);
1048
1049         if (!test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
1050                 /* If only reply is a RST, we can consider ourselves not to
1051                    have an established connection: this is a fairly common
1052                    problem case, so we can delete the conntrack
1053                    immediately.  --RR */
1054                 if (th->rst) {
1055                         nf_ct_kill_acct(ct, ctinfo, skb);
1056                         return NF_ACCEPT;
1057                 }
1058                 /* ESTABLISHED without SEEN_REPLY, i.e. mid-connection
1059                  * pickup with loose=1. Avoid large ESTABLISHED timeout.
1060                  */
1061                 if (new_state == TCP_CONNTRACK_ESTABLISHED &&
1062                     timeout > timeouts[TCP_CONNTRACK_UNACK])
1063                         timeout = timeouts[TCP_CONNTRACK_UNACK];
1064         } else if (!test_bit(IPS_ASSURED_BIT, &ct->status)
1065                    && (old_state == TCP_CONNTRACK_SYN_RECV
1066                        || old_state == TCP_CONNTRACK_ESTABLISHED)
1067                    && new_state == TCP_CONNTRACK_ESTABLISHED) {
1068                 /* Set ASSURED if we see see valid ack in ESTABLISHED
1069                    after SYN_RECV or a valid answer for a picked up
1070                    connection. */
1071                 set_bit(IPS_ASSURED_BIT, &ct->status);
1072                 nf_conntrack_event_cache(IPCT_ASSURED, ct);
1073         }
1074         nf_ct_refresh_acct(ct, ctinfo, skb, timeout);
1075
1076         return NF_ACCEPT;
1077 }
1078
1079 /* Called when a new connection for this protocol found. */
1080 static bool tcp_new(struct nf_conn *ct, const struct sk_buff *skb,
1081                     unsigned int dataoff, unsigned int *timeouts)
1082 {
1083         enum tcp_conntrack new_state;
1084         const struct tcphdr *th;
1085         struct tcphdr _tcph;
1086         struct net *net = nf_ct_net(ct);
1087         struct nf_tcp_net *tn = tcp_pernet(net);
1088         const struct ip_ct_tcp_state *sender = &ct->proto.tcp.seen[0];
1089         const struct ip_ct_tcp_state *receiver = &ct->proto.tcp.seen[1];
1090
1091         th = skb_header_pointer(skb, dataoff, sizeof(_tcph), &_tcph);
1092         BUG_ON(th == NULL);
1093
1094         /* Don't need lock here: this conntrack not in circulation yet */
1095         new_state = tcp_conntracks[0][get_conntrack_index(th)][TCP_CONNTRACK_NONE];
1096
1097         /* Invalid: delete conntrack */
1098         if (new_state >= TCP_CONNTRACK_MAX) {
1099                 pr_debug("nf_ct_tcp: invalid new deleting.\n");
1100                 return false;
1101         }
1102
1103         if (new_state == TCP_CONNTRACK_SYN_SENT) {
1104                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1105                 /* SYN packet */
1106                 ct->proto.tcp.seen[0].td_end =
1107                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1108                                              dataoff, th);
1109                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1110                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1111                         ct->proto.tcp.seen[0].td_maxwin = 1;
1112                 ct->proto.tcp.seen[0].td_maxend =
1113                         ct->proto.tcp.seen[0].td_end;
1114
1115                 tcp_options(skb, dataoff, th, &ct->proto.tcp.seen[0]);
1116         } else if (tn->tcp_loose == 0) {
1117                 /* Don't try to pick up connections. */
1118                 return false;
1119         } else {
1120                 memset(&ct->proto.tcp, 0, sizeof(ct->proto.tcp));
1121                 /*
1122                  * We are in the middle of a connection,
1123                  * its history is lost for us.
1124                  * Let's try to use the data from the packet.
1125                  */
1126                 ct->proto.tcp.seen[0].td_end =
1127                         segment_seq_plus_len(ntohl(th->seq), skb->len,
1128                                              dataoff, th);
1129                 ct->proto.tcp.seen[0].td_maxwin = ntohs(th->window);
1130                 if (ct->proto.tcp.seen[0].td_maxwin == 0)
1131                         ct->proto.tcp.seen[0].td_maxwin = 1;
1132                 ct->proto.tcp.seen[0].td_maxend =
1133                         ct->proto.tcp.seen[0].td_end +
1134                         ct->proto.tcp.seen[0].td_maxwin;
1135
1136                 /* We assume SACK and liberal window checking to handle
1137                  * window scaling */
1138                 ct->proto.tcp.seen[0].flags =
1139                 ct->proto.tcp.seen[1].flags = IP_CT_TCP_FLAG_SACK_PERM |
1140                                               IP_CT_TCP_FLAG_BE_LIBERAL;
1141         }
1142
1143         /* tcp_packet will set them */
1144         ct->proto.tcp.last_index = TCP_NONE_SET;
1145
1146         pr_debug("tcp_new: sender end=%u maxend=%u maxwin=%u scale=%i "
1147                  "receiver end=%u maxend=%u maxwin=%u scale=%i\n",
1148                  sender->td_end, sender->td_maxend, sender->td_maxwin,
1149                  sender->td_scale,
1150                  receiver->td_end, receiver->td_maxend, receiver->td_maxwin,
1151                  receiver->td_scale);
1152         return true;
1153 }
1154
1155 static bool tcp_can_early_drop(const struct nf_conn *ct)
1156 {
1157         switch (ct->proto.tcp.state) {
1158         case TCP_CONNTRACK_FIN_WAIT:
1159         case TCP_CONNTRACK_LAST_ACK:
1160         case TCP_CONNTRACK_TIME_WAIT:
1161         case TCP_CONNTRACK_CLOSE:
1162         case TCP_CONNTRACK_CLOSE_WAIT:
1163                 return true;
1164         default:
1165                 break;
1166         }
1167
1168         return false;
1169 }
1170
1171 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1172
1173 #include <linux/netfilter/nfnetlink.h>
1174 #include <linux/netfilter/nfnetlink_conntrack.h>
1175
1176 static int tcp_to_nlattr(struct sk_buff *skb, struct nlattr *nla,
1177                          struct nf_conn *ct)
1178 {
1179         struct nlattr *nest_parms;
1180         struct nf_ct_tcp_flags tmp = {};
1181
1182         spin_lock_bh(&ct->lock);
1183         nest_parms = nla_nest_start(skb, CTA_PROTOINFO_TCP | NLA_F_NESTED);
1184         if (!nest_parms)
1185                 goto nla_put_failure;
1186
1187         if (nla_put_u8(skb, CTA_PROTOINFO_TCP_STATE, ct->proto.tcp.state) ||
1188             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_ORIGINAL,
1189                        ct->proto.tcp.seen[0].td_scale) ||
1190             nla_put_u8(skb, CTA_PROTOINFO_TCP_WSCALE_REPLY,
1191                        ct->proto.tcp.seen[1].td_scale))
1192                 goto nla_put_failure;
1193
1194         tmp.flags = ct->proto.tcp.seen[0].flags;
1195         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_ORIGINAL,
1196                     sizeof(struct nf_ct_tcp_flags), &tmp))
1197                 goto nla_put_failure;
1198
1199         tmp.flags = ct->proto.tcp.seen[1].flags;
1200         if (nla_put(skb, CTA_PROTOINFO_TCP_FLAGS_REPLY,
1201                     sizeof(struct nf_ct_tcp_flags), &tmp))
1202                 goto nla_put_failure;
1203         spin_unlock_bh(&ct->lock);
1204
1205         nla_nest_end(skb, nest_parms);
1206
1207         return 0;
1208
1209 nla_put_failure:
1210         spin_unlock_bh(&ct->lock);
1211         return -1;
1212 }
1213
1214 static const struct nla_policy tcp_nla_policy[CTA_PROTOINFO_TCP_MAX+1] = {
1215         [CTA_PROTOINFO_TCP_STATE]           = { .type = NLA_U8 },
1216         [CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] = { .type = NLA_U8 },
1217         [CTA_PROTOINFO_TCP_WSCALE_REPLY]    = { .type = NLA_U8 },
1218         [CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]  = { .len = sizeof(struct nf_ct_tcp_flags) },
1219         [CTA_PROTOINFO_TCP_FLAGS_REPLY]     = { .len =  sizeof(struct nf_ct_tcp_flags) },
1220 };
1221
1222 static int nlattr_to_tcp(struct nlattr *cda[], struct nf_conn *ct)
1223 {
1224         struct nlattr *pattr = cda[CTA_PROTOINFO_TCP];
1225         struct nlattr *tb[CTA_PROTOINFO_TCP_MAX+1];
1226         int err;
1227
1228         /* updates could not contain anything about the private
1229          * protocol info, in that case skip the parsing */
1230         if (!pattr)
1231                 return 0;
1232
1233         err = nla_parse_nested(tb, CTA_PROTOINFO_TCP_MAX, pattr,
1234                                tcp_nla_policy, NULL);
1235         if (err < 0)
1236                 return err;
1237
1238         if (tb[CTA_PROTOINFO_TCP_STATE] &&
1239             nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]) >= TCP_CONNTRACK_MAX)
1240                 return -EINVAL;
1241
1242         spin_lock_bh(&ct->lock);
1243         if (tb[CTA_PROTOINFO_TCP_STATE])
1244                 ct->proto.tcp.state = nla_get_u8(tb[CTA_PROTOINFO_TCP_STATE]);
1245
1246         if (tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]) {
1247                 struct nf_ct_tcp_flags *attr =
1248                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_ORIGINAL]);
1249                 ct->proto.tcp.seen[0].flags &= ~attr->mask;
1250                 ct->proto.tcp.seen[0].flags |= attr->flags & attr->mask;
1251         }
1252
1253         if (tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]) {
1254                 struct nf_ct_tcp_flags *attr =
1255                         nla_data(tb[CTA_PROTOINFO_TCP_FLAGS_REPLY]);
1256                 ct->proto.tcp.seen[1].flags &= ~attr->mask;
1257                 ct->proto.tcp.seen[1].flags |= attr->flags & attr->mask;
1258         }
1259
1260         if (tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL] &&
1261             tb[CTA_PROTOINFO_TCP_WSCALE_REPLY] &&
1262             ct->proto.tcp.seen[0].flags & IP_CT_TCP_FLAG_WINDOW_SCALE &&
1263             ct->proto.tcp.seen[1].flags & IP_CT_TCP_FLAG_WINDOW_SCALE) {
1264                 ct->proto.tcp.seen[0].td_scale =
1265                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_ORIGINAL]);
1266                 ct->proto.tcp.seen[1].td_scale =
1267                         nla_get_u8(tb[CTA_PROTOINFO_TCP_WSCALE_REPLY]);
1268         }
1269         spin_unlock_bh(&ct->lock);
1270
1271         return 0;
1272 }
1273
1274 static int tcp_nlattr_size(void)
1275 {
1276         return nla_total_size(0)           /* CTA_PROTOINFO_TCP */
1277                 + nla_policy_len(tcp_nla_policy, CTA_PROTOINFO_TCP_MAX + 1);
1278 }
1279
1280 static unsigned int tcp_nlattr_tuple_size(void)
1281 {
1282         static unsigned int size __read_mostly;
1283
1284         if (!size)
1285                 size = nla_policy_len(nf_ct_port_nla_policy, CTA_PROTO_MAX + 1);
1286
1287         return size;
1288 }
1289 #endif
1290
1291 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1292
1293 #include <linux/netfilter/nfnetlink.h>
1294 #include <linux/netfilter/nfnetlink_cttimeout.h>
1295
1296 static int tcp_timeout_nlattr_to_obj(struct nlattr *tb[],
1297                                      struct net *net, void *data)
1298 {
1299         unsigned int *timeouts = data;
1300         struct nf_tcp_net *tn = tcp_pernet(net);
1301         int i;
1302
1303         /* set default TCP timeouts. */
1304         for (i=0; i<TCP_CONNTRACK_TIMEOUT_MAX; i++)
1305                 timeouts[i] = tn->timeouts[i];
1306
1307         if (tb[CTA_TIMEOUT_TCP_SYN_SENT]) {
1308                 timeouts[TCP_CONNTRACK_SYN_SENT] =
1309                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT]))*HZ;
1310         }
1311         if (tb[CTA_TIMEOUT_TCP_SYN_RECV]) {
1312                 timeouts[TCP_CONNTRACK_SYN_RECV] =
1313                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_RECV]))*HZ;
1314         }
1315         if (tb[CTA_TIMEOUT_TCP_ESTABLISHED]) {
1316                 timeouts[TCP_CONNTRACK_ESTABLISHED] =
1317                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_ESTABLISHED]))*HZ;
1318         }
1319         if (tb[CTA_TIMEOUT_TCP_FIN_WAIT]) {
1320                 timeouts[TCP_CONNTRACK_FIN_WAIT] =
1321                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_FIN_WAIT]))*HZ;
1322         }
1323         if (tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]) {
1324                 timeouts[TCP_CONNTRACK_CLOSE_WAIT] =
1325                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE_WAIT]))*HZ;
1326         }
1327         if (tb[CTA_TIMEOUT_TCP_LAST_ACK]) {
1328                 timeouts[TCP_CONNTRACK_LAST_ACK] =
1329                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_LAST_ACK]))*HZ;
1330         }
1331         if (tb[CTA_TIMEOUT_TCP_TIME_WAIT]) {
1332                 timeouts[TCP_CONNTRACK_TIME_WAIT] =
1333                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_TIME_WAIT]))*HZ;
1334         }
1335         if (tb[CTA_TIMEOUT_TCP_CLOSE]) {
1336                 timeouts[TCP_CONNTRACK_CLOSE] =
1337                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_CLOSE]))*HZ;
1338         }
1339         if (tb[CTA_TIMEOUT_TCP_SYN_SENT2]) {
1340                 timeouts[TCP_CONNTRACK_SYN_SENT2] =
1341                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_SYN_SENT2]))*HZ;
1342         }
1343         if (tb[CTA_TIMEOUT_TCP_RETRANS]) {
1344                 timeouts[TCP_CONNTRACK_RETRANS] =
1345                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_RETRANS]))*HZ;
1346         }
1347         if (tb[CTA_TIMEOUT_TCP_UNACK]) {
1348                 timeouts[TCP_CONNTRACK_UNACK] =
1349                         ntohl(nla_get_be32(tb[CTA_TIMEOUT_TCP_UNACK]))*HZ;
1350         }
1351         return 0;
1352 }
1353
1354 static int
1355 tcp_timeout_obj_to_nlattr(struct sk_buff *skb, const void *data)
1356 {
1357         const unsigned int *timeouts = data;
1358
1359         if (nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT,
1360                         htonl(timeouts[TCP_CONNTRACK_SYN_SENT] / HZ)) ||
1361             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_RECV,
1362                          htonl(timeouts[TCP_CONNTRACK_SYN_RECV] / HZ)) ||
1363             nla_put_be32(skb, CTA_TIMEOUT_TCP_ESTABLISHED,
1364                          htonl(timeouts[TCP_CONNTRACK_ESTABLISHED] / HZ)) ||
1365             nla_put_be32(skb, CTA_TIMEOUT_TCP_FIN_WAIT,
1366                          htonl(timeouts[TCP_CONNTRACK_FIN_WAIT] / HZ)) ||
1367             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE_WAIT,
1368                          htonl(timeouts[TCP_CONNTRACK_CLOSE_WAIT] / HZ)) ||
1369             nla_put_be32(skb, CTA_TIMEOUT_TCP_LAST_ACK,
1370                          htonl(timeouts[TCP_CONNTRACK_LAST_ACK] / HZ)) ||
1371             nla_put_be32(skb, CTA_TIMEOUT_TCP_TIME_WAIT,
1372                          htonl(timeouts[TCP_CONNTRACK_TIME_WAIT] / HZ)) ||
1373             nla_put_be32(skb, CTA_TIMEOUT_TCP_CLOSE,
1374                          htonl(timeouts[TCP_CONNTRACK_CLOSE] / HZ)) ||
1375             nla_put_be32(skb, CTA_TIMEOUT_TCP_SYN_SENT2,
1376                          htonl(timeouts[TCP_CONNTRACK_SYN_SENT2] / HZ)) ||
1377             nla_put_be32(skb, CTA_TIMEOUT_TCP_RETRANS,
1378                          htonl(timeouts[TCP_CONNTRACK_RETRANS] / HZ)) ||
1379             nla_put_be32(skb, CTA_TIMEOUT_TCP_UNACK,
1380                          htonl(timeouts[TCP_CONNTRACK_UNACK] / HZ)))
1381                 goto nla_put_failure;
1382         return 0;
1383
1384 nla_put_failure:
1385         return -ENOSPC;
1386 }
1387
1388 static const struct nla_policy tcp_timeout_nla_policy[CTA_TIMEOUT_TCP_MAX+1] = {
1389         [CTA_TIMEOUT_TCP_SYN_SENT]      = { .type = NLA_U32 },
1390         [CTA_TIMEOUT_TCP_SYN_RECV]      = { .type = NLA_U32 },
1391         [CTA_TIMEOUT_TCP_ESTABLISHED]   = { .type = NLA_U32 },
1392         [CTA_TIMEOUT_TCP_FIN_WAIT]      = { .type = NLA_U32 },
1393         [CTA_TIMEOUT_TCP_CLOSE_WAIT]    = { .type = NLA_U32 },
1394         [CTA_TIMEOUT_TCP_LAST_ACK]      = { .type = NLA_U32 },
1395         [CTA_TIMEOUT_TCP_TIME_WAIT]     = { .type = NLA_U32 },
1396         [CTA_TIMEOUT_TCP_CLOSE]         = { .type = NLA_U32 },
1397         [CTA_TIMEOUT_TCP_SYN_SENT2]     = { .type = NLA_U32 },
1398         [CTA_TIMEOUT_TCP_RETRANS]       = { .type = NLA_U32 },
1399         [CTA_TIMEOUT_TCP_UNACK]         = { .type = NLA_U32 },
1400 };
1401 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1402
1403 #ifdef CONFIG_SYSCTL
1404 static struct ctl_table tcp_sysctl_table[] = {
1405         {
1406                 .procname       = "nf_conntrack_tcp_timeout_syn_sent",
1407                 .maxlen         = sizeof(unsigned int),
1408                 .mode           = 0644,
1409                 .proc_handler   = proc_dointvec_jiffies,
1410         },
1411         {
1412                 .procname       = "nf_conntrack_tcp_timeout_syn_recv",
1413                 .maxlen         = sizeof(unsigned int),
1414                 .mode           = 0644,
1415                 .proc_handler   = proc_dointvec_jiffies,
1416         },
1417         {
1418                 .procname       = "nf_conntrack_tcp_timeout_established",
1419                 .maxlen         = sizeof(unsigned int),
1420                 .mode           = 0644,
1421                 .proc_handler   = proc_dointvec_jiffies,
1422         },
1423         {
1424                 .procname       = "nf_conntrack_tcp_timeout_fin_wait",
1425                 .maxlen         = sizeof(unsigned int),
1426                 .mode           = 0644,
1427                 .proc_handler   = proc_dointvec_jiffies,
1428         },
1429         {
1430                 .procname       = "nf_conntrack_tcp_timeout_close_wait",
1431                 .maxlen         = sizeof(unsigned int),
1432                 .mode           = 0644,
1433                 .proc_handler   = proc_dointvec_jiffies,
1434         },
1435         {
1436                 .procname       = "nf_conntrack_tcp_timeout_last_ack",
1437                 .maxlen         = sizeof(unsigned int),
1438                 .mode           = 0644,
1439                 .proc_handler   = proc_dointvec_jiffies,
1440         },
1441         {
1442                 .procname       = "nf_conntrack_tcp_timeout_time_wait",
1443                 .maxlen         = sizeof(unsigned int),
1444                 .mode           = 0644,
1445                 .proc_handler   = proc_dointvec_jiffies,
1446         },
1447         {
1448                 .procname       = "nf_conntrack_tcp_timeout_close",
1449                 .maxlen         = sizeof(unsigned int),
1450                 .mode           = 0644,
1451                 .proc_handler   = proc_dointvec_jiffies,
1452         },
1453         {
1454                 .procname       = "nf_conntrack_tcp_timeout_max_retrans",
1455                 .maxlen         = sizeof(unsigned int),
1456                 .mode           = 0644,
1457                 .proc_handler   = proc_dointvec_jiffies,
1458         },
1459         {
1460                 .procname       = "nf_conntrack_tcp_timeout_unacknowledged",
1461                 .maxlen         = sizeof(unsigned int),
1462                 .mode           = 0644,
1463                 .proc_handler   = proc_dointvec_jiffies,
1464         },
1465         {
1466                 .procname       = "nf_conntrack_tcp_loose",
1467                 .maxlen         = sizeof(unsigned int),
1468                 .mode           = 0644,
1469                 .proc_handler   = proc_dointvec,
1470         },
1471         {
1472                 .procname       = "nf_conntrack_tcp_be_liberal",
1473                 .maxlen         = sizeof(unsigned int),
1474                 .mode           = 0644,
1475                 .proc_handler   = proc_dointvec,
1476         },
1477         {
1478                 .procname       = "nf_conntrack_tcp_max_retrans",
1479                 .maxlen         = sizeof(unsigned int),
1480                 .mode           = 0644,
1481                 .proc_handler   = proc_dointvec,
1482         },
1483         { }
1484 };
1485 #endif /* CONFIG_SYSCTL */
1486
1487 static int tcp_kmemdup_sysctl_table(struct nf_proto_net *pn,
1488                                     struct nf_tcp_net *tn)
1489 {
1490 #ifdef CONFIG_SYSCTL
1491         if (pn->ctl_table)
1492                 return 0;
1493
1494         pn->ctl_table = kmemdup(tcp_sysctl_table,
1495                                 sizeof(tcp_sysctl_table),
1496                                 GFP_KERNEL);
1497         if (!pn->ctl_table)
1498                 return -ENOMEM;
1499
1500         pn->ctl_table[0].data = &tn->timeouts[TCP_CONNTRACK_SYN_SENT];
1501         pn->ctl_table[1].data = &tn->timeouts[TCP_CONNTRACK_SYN_RECV];
1502         pn->ctl_table[2].data = &tn->timeouts[TCP_CONNTRACK_ESTABLISHED];
1503         pn->ctl_table[3].data = &tn->timeouts[TCP_CONNTRACK_FIN_WAIT];
1504         pn->ctl_table[4].data = &tn->timeouts[TCP_CONNTRACK_CLOSE_WAIT];
1505         pn->ctl_table[5].data = &tn->timeouts[TCP_CONNTRACK_LAST_ACK];
1506         pn->ctl_table[6].data = &tn->timeouts[TCP_CONNTRACK_TIME_WAIT];
1507         pn->ctl_table[7].data = &tn->timeouts[TCP_CONNTRACK_CLOSE];
1508         pn->ctl_table[8].data = &tn->timeouts[TCP_CONNTRACK_RETRANS];
1509         pn->ctl_table[9].data = &tn->timeouts[TCP_CONNTRACK_UNACK];
1510         pn->ctl_table[10].data = &tn->tcp_loose;
1511         pn->ctl_table[11].data = &tn->tcp_be_liberal;
1512         pn->ctl_table[12].data = &tn->tcp_max_retrans;
1513 #endif
1514         return 0;
1515 }
1516
1517 static int tcp_init_net(struct net *net, u_int16_t proto)
1518 {
1519         struct nf_tcp_net *tn = tcp_pernet(net);
1520         struct nf_proto_net *pn = &tn->pn;
1521
1522         if (!pn->users) {
1523                 int i;
1524
1525                 for (i = 0; i < TCP_CONNTRACK_TIMEOUT_MAX; i++)
1526                         tn->timeouts[i] = tcp_timeouts[i];
1527
1528                 tn->tcp_loose = nf_ct_tcp_loose;
1529                 tn->tcp_be_liberal = nf_ct_tcp_be_liberal;
1530                 tn->tcp_max_retrans = nf_ct_tcp_max_retrans;
1531         }
1532
1533         return tcp_kmemdup_sysctl_table(pn, tn);
1534 }
1535
1536 static struct nf_proto_net *tcp_get_net_proto(struct net *net)
1537 {
1538         return &net->ct.nf_ct_proto.tcp.pn;
1539 }
1540
1541 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp4 __read_mostly =
1542 {
1543         .l3proto                = PF_INET,
1544         .l4proto                = IPPROTO_TCP,
1545         .pkt_to_tuple           = tcp_pkt_to_tuple,
1546         .invert_tuple           = tcp_invert_tuple,
1547 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1548         .print_conntrack        = tcp_print_conntrack,
1549 #endif
1550         .packet                 = tcp_packet,
1551         .get_timeouts           = tcp_get_timeouts,
1552         .new                    = tcp_new,
1553         .error                  = tcp_error,
1554         .can_early_drop         = tcp_can_early_drop,
1555 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1556         .to_nlattr              = tcp_to_nlattr,
1557         .nlattr_size            = tcp_nlattr_size,
1558         .from_nlattr            = nlattr_to_tcp,
1559         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1560         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1561         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1562         .nla_policy             = nf_ct_port_nla_policy,
1563 #endif
1564 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1565         .ctnl_timeout           = {
1566                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1567                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1568                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1569                 .obj_size       = sizeof(unsigned int) *
1570                                         TCP_CONNTRACK_TIMEOUT_MAX,
1571                 .nla_policy     = tcp_timeout_nla_policy,
1572         },
1573 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1574         .init_net               = tcp_init_net,
1575         .get_net_proto          = tcp_get_net_proto,
1576 };
1577 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp4);
1578
1579 struct nf_conntrack_l4proto nf_conntrack_l4proto_tcp6 __read_mostly =
1580 {
1581         .l3proto                = PF_INET6,
1582         .l4proto                = IPPROTO_TCP,
1583         .pkt_to_tuple           = tcp_pkt_to_tuple,
1584         .invert_tuple           = tcp_invert_tuple,
1585 #ifdef CONFIG_NF_CONNTRACK_PROCFS
1586         .print_conntrack        = tcp_print_conntrack,
1587 #endif
1588         .packet                 = tcp_packet,
1589         .get_timeouts           = tcp_get_timeouts,
1590         .new                    = tcp_new,
1591         .error                  = tcp_error,
1592         .can_early_drop         = tcp_can_early_drop,
1593 #if IS_ENABLED(CONFIG_NF_CT_NETLINK)
1594         .to_nlattr              = tcp_to_nlattr,
1595         .nlattr_size            = tcp_nlattr_size,
1596         .from_nlattr            = nlattr_to_tcp,
1597         .tuple_to_nlattr        = nf_ct_port_tuple_to_nlattr,
1598         .nlattr_to_tuple        = nf_ct_port_nlattr_to_tuple,
1599         .nlattr_tuple_size      = tcp_nlattr_tuple_size,
1600         .nla_policy             = nf_ct_port_nla_policy,
1601 #endif
1602 #if IS_ENABLED(CONFIG_NF_CT_NETLINK_TIMEOUT)
1603         .ctnl_timeout           = {
1604                 .nlattr_to_obj  = tcp_timeout_nlattr_to_obj,
1605                 .obj_to_nlattr  = tcp_timeout_obj_to_nlattr,
1606                 .nlattr_max     = CTA_TIMEOUT_TCP_MAX,
1607                 .obj_size       = sizeof(unsigned int) *
1608                                         TCP_CONNTRACK_TIMEOUT_MAX,
1609                 .nla_policy     = tcp_timeout_nla_policy,
1610         },
1611 #endif /* CONFIG_NF_CT_NETLINK_TIMEOUT */
1612         .init_net               = tcp_init_net,
1613         .get_net_proto          = tcp_get_net_proto,
1614 };
1615 EXPORT_SYMBOL_GPL(nf_conntrack_l4proto_tcp6);